Jurassic 2.0.4 Release Notes

We are excited to announce the release of Jurassic version 2.0.4, now in production. The new version is automatically available to all users, giving you access to all new features and improvements without any action needed. For users on Amazon SageMaker, you can enjoy the benefits of this update by specifying a new ARN. This version introduces a major upgrade to model safety by effectively declining responses to prompts that may provoke any unwanted behavior such as sexual, violent or racist content. Besides the major safety improvements, this release also includes significant advancements in the model’s ability to respond to legitimate prompts. The following is a list of new features that enhance the usability of our models.


Jurassic-2 Model Renaming: No Action Needed Thanks to Automatic Rerouting

We're happy to share some exciting updates about our Jurassic models at AI21 Studio. In our quest to make our services as user-friendly and intuitive as possible, we have made some changes to the names of our models. However, we want to reassure you that no immediate action is required on your part as we have implemented an automatic rerouting system to ensure uninterrupted service. Here are the details:


Contexto más largo in Contextual-Answers API

Contextual Answers API [Beta] now supports an expanded context of 50,000 characters, allowing more text to be included in your source document. In addition, the API provides better support for multilingual capabilities, including Spanish, French, German, Portuguese, Italian and Dutch.

Transitioning to Jurassic-2: Important Updates on Model Retirements and Pricing Changes

We have a few updates to share regarding our Jurassic models at AI21 Studio:


Extending Context Window: Jurassic-2 Instruct Models Now Support 8K Tokens

We're thrilled to announce the extension of the 8K tokens support to our Jurassic-2 Instruct models, specifically J2-Grande-Instruct and J2-Jumbo-Instruct. These models, which are designed to meticulously follow instructions, can now accommodate a context window of 8,192 tokens, allowing for even more detailed and extensive interactions.


Fine-tuning Jurassic-2 models is now available!

Custom model training on top of J2-Grande and J2-Large is now available.


Increased Context Length for J2 Foundation Models

We are excited to announce that our Jurassic-2 Foundation models, including J2-Large, J2-Grande, and J2-Jumbo, now support an increased context length of up to 8K tokens (8192 tokens to be exact). This enhancement allows users to work with longer prompts and receive more comprehensive completions.


Contextual Answers API [BETA] Released

We are pleased to announce the release of the Contextual Answers API, a powerful tool for providing accurate and reliable question answering based on specific document context. This API is designed to avoid any factual issues that may arise when using language models to provide answers, and it receives document text and a question and returns an answer based solely on the provided context.


Introducing Wordtune-Powered Task-Specific Models

We're thrilled to introduce our latest offering - the Task-Specific Models package. This suite includes specialized language processing capabilities, making it effortless for you to integrate cutting-edge features into your systems. Our APIs are optimized and fine-tuned to ensure fast response times and high-quality results. We're proud to announce that all of our Task-Specific Models are supported through our Python SDK, offering easy integration with your existing Python-based systems. These models are powered by the same engines behind our popular Wordtune products and include: